This PR updates the score.py script of the SUM module so that both bert_score and rouge provide scores when the source (src) documents are given as the target and the machine generated summaries are given as the predictions. These new scores are tagged as *_src_hyp and *_hypo_ref, following the conventions in the original BARTScore repo.
One interesting note: bert_score is the same regardless of the order of inputs (e.g. (src, hypo) <==> (hypo, src)) but rouge is not.
This PR updates the
score.pyscript of theSUMmodule so that bothbert_scoreandrougeprovide scores when the source (src) documents are given as the target and the machine generated summaries are given as the predictions. These new scores are tagged as*_src_hypand*_hypo_ref, following the conventions in the original BARTScore repo.One interesting note:
bert_scoreis the same regardless of the order of inputs (e.g.(src, hypo)<==>(hypo, src)) butrougeis not.